A86: Re: Re: Fast Random Number Generation

[Prev][Next][Index][Thread]

A86: Re: Re: Fast Random Number Generation

Mine doesn't return the same number everytime!  It uses the R register to
grab a byte from the ROM page, and then it uses that to grab another byte.
If you run the program, you'll see that the same numbers keep wanting to pop
up, but not in order, and the distribution seems pretty good.  But now that
I think of it, you only need to read one byte, since that will always be the
same for the second byte.  Other routines use seeds, so...

I changed the routine to use seeds (self modifying code), and it works much
better now I believe.  Attached is the screenshot of a histogram.  I had it
make 1000 numbers and store it in a list, then did a plot.  They seem to be
heavy on the right side (towards $ff).

;===========================================
; RandomNumber     [Assembly Coder's Zenith]
; out: A = psuedo random number
; modified: AF, BC    (14 bytes/62 t-states)
;===========================================
RandomNumber:
      ld a,r
  ld b,a
  ld a,($1234)
  _@RandomNumberSeed =$-2
  ld c,a
  ld a,(bc)
  ld (_@RandomNumberSeed),a
  xor b
  sub b
  add a,c
  ret

P.S. If anyone can explain why Outlook Express 5 acts weird when you paste
text, please let me know.  =)  (email me and not the list!)

----- Original Message -----
From: Justin Karneges <jgkarneges@ucdavis.edu>
To: <assembly-86@lists.ticalc.org>
Sent: Wednesday, February 24, 1999 6:44 PM
Subject: A86: Re: Fast Random Number Generation


>
>A similar type of random number routine was used for the TI-85 version of
>Joltima, but it screwed the game up badly and was replaced with Hideaki
>Omuro's (the Crash guy) routine which works very well by using
>self-modifying code.  The reason this style of routine didn't work with
>Joltima is this:
>
>When you're walking around, I had it keep getting a random number for each
>step you take.  If the result was a 1, then the battle began.  Otherwise,
>you kept walking.  Once a battle began, another random number was requested
>to decide who you fight.  Now since using the R register and/or many other
>'quick' random number methods gives you 'pattern' stylish random numbers,
>Joltima would always get the same number after the 1.  See what I mean?  It
>would keep getting numbers which are not 1's, and then someday it would
>finally get a 1.  Immediately following the 1 was the same number
everytime.
>The result?  In the unfixed Joltima85, you fought the same enemies in every
>battle!  YUCK!  I still got great e-mail comments from TI-85 players, but
>they were sure missing out.  Imagine fighting the same battle while going
>through one of the caves!  Anyways...
>
>I mention this because I'm sure many potential RPG's and other type games
>would use random numbers in this fashion.  To prevent game-destroying
>patterns, use the built-in ROM routine.  It's VERY good, but slow.  In
>David's case, he needs a fast random number so I suggest using Hideaki's
>routine.  It's extremely quick.  I would put it in this post, but I don't
>know if he wants it out.  I guess that was kind of dumb of me to suggest
>using some code that isn't even available.  Go download the CrASH v1.5
shell
>for the TI-82 and look at the readme file to get his email address and
>contact him.
>
>-Justin Karneges [Infiniti]
>
>>I've been doing some experimenting to come up with a quick routine for
>>random number generation.  The routine below (47t without ret) is what I
>>have come up with.  The results are semi-random.  The same numbers keep
>>coming up, as you'd expect considering it's pulling opcodes off rom pages,
>>though the distribution seems fairly decent.  For my purposes, the actual
>>number doesn't matter as long as the distribution works.  Attached is the
>>skeleton program used to test the routine.  It just prints out six rows of
>>random numbers in hex.
>>
>> Speed is important, I don't want it to take longer to generate a random
>>number than it would to draw a sprite, as is the case with Mardell's
>>generator from ZTetris when using a large numbers for A.  If speed didn't
>>matter, I'd use his as it makes much better numbers than mine ;-)
>>
>>Any suggestions/improvements?
>>
>>; returns: a = psuedo random number
>>RandomNumber:
>>     ld a,r
>>   ld b,a
>>  add a,b
>>  ld c,a
>>  ld a,(bc)
>>  ld c,a
>>  ld a,(bc)
>>  xor b
>>  add a,c
>>  ret
>>
>>--
>>David Phillips <david@acz.org>
>>http://www.acz.org/
>>
>
>
>

randhist.tif